Risk Scoring and Mapping Are Reshaping Property Insurance

Historical Background—Scoring

As a rookie independent agent, I was not accustomed to using the term "score" when writing personal lines property policies. Instead, it was essential to specify key underwriting criteria to quote homeowners policies properly. For example, it was critical to know the building's construction, age, roof type, recent upgrades, loss history, and the proper fire protection code. Using this information, it was common for agents to "manually" quote the policy using company manuals, as the widespread use of computers in the process was not yet commonplace.

Eventually, this process was facilitated using computers and agent rating programs. Even though I was an agent in a hail-prone state, natural catastrophes did not usually cause upfront rating issues. They did occasionally cause loss ratio issues with our insurance company partners.

The first introduction to scoring was credit scores. Although this was more focused on personal auto issues, it also applied to homeowners policies, since it was common to cross-sell these two policies with the same insurer to offer discounts to customers. As they were introduced, agents and their customers wondered what a person's credit score had to do with their insurability. At some point, the developers convinced the regulators and front-end users that there was a statistical relationship between a person's credit score and the likelihood of losses. From the onset, this created the following two issues for insurance agents.

• Should agents have access to their customers' credit scores since this creates an invasion of privacy?
• Could agents counsel their customers on repairing their credit and/or fixing errors in their credit reports to help them get better premiums?

The solutions to these questions were quite simple. The agent had to disclose to their customers that the policy rating would include a credit score check and that the actual scores would be shared with the companies for final rating, but the agents would not have access to these credit scores. If an adverse credit score caused an increase in the premium, a separate communication would be sent to the customer indicating that they needed to reach out to the credit score company to identify the specifics on how this adverse credit score was computed and to determine the method to correct any erroneous scoring information.

Was this credit scoring system fair? Like most such questions, it depends on who you ask. The most difficult conversation for agents was explaining the credit and insurance relationship to customers who either refused to sign the privacy and credit use agreements or who learned of an adverse credit decision. A key to the use of this score was the individual's ability to improve their credit score in most cases. Like any automated system collecting data, some situations could be easily resolved by providing the correct data to the credit rating company, although this process was very time-consuming for the public, particularly in the days before widespread computer access and the correction of this information.

Bottom line: This system gained wide acceptance after some initial issues and continues to be used as an underwriting criterion in present-day personal lines rating. Curiously, there can be differences between a customer's credit score and their insurance credit score due to proprietary algorithms used to compute these scores.

Historical Background—Risk Mapping

Before the extensive use of computers in personal lines rating, all mapping was manual. In some cases, agents or companies used specific maps to identify undesirable risks. A common term for these maps was "redlining," which regulators deemed discriminatory and therefore illegal. Typically, these maps drew red lines around zip codes or neighborhoods to outlaw their use. In some situations, insurance companies could apply for higher rates in specific geographic regions if they could provide credible data showing increased exposure that warranted additional rates.

The most extensive traditional use of risk mapping is flood mapping. These maps were created by the National Flood Insurance Program (NFIP) on paper to identify, primarily based on historical and topographic data, properties exposed to the flood peril. These maps are now available digitally and still serve their original purposes. However, changes in earth's weather patterns are making these maps more challenging to use for identifying flood-exposed properties.

Traditionally, risk mapping used paper maps to display information before the extensive use of computers, which made these maps very difficult to maintain. As the capabilities of larger computers with greater access to data emerged, it was inevitable to use this new computing power to measure the risks faced by insurance consumers in the 21st century, initially focusing on their exposure to natural disasters.

Risk Scoring Versus Risk Mapping

While similar, there is a fundamental difference between these two terms. Risk scoring is a method for quantifying a property's exposure to loss by giving it a numerical value. Generally, the higher the score, the higher the risk. Higher scores would lead an underwriter to decline to write the property or to surcharge the premium, recognizing its higher loss exposure.

Typically, these risk scores are computed by third-party vendors who use proprietary algorithms to quantify the potential risk of individual properties. They obtain data from many sources to calculate these scores, but they usually rely heavily on satellite imagery to identify specific properties, along with other data. These scores are combined with the traditional personal property underwriting criteria, credit scores, and insurer-specific actuarial data to make underwriting decisions.

Risk mapping is usually an aggregation of risk scores displayed in a geospatial format, showing emergency, risk, or underwriting managers the relative risk in specific geographic areas. For example, suppose a neighborhood or political entity is determined to have high wildfire exposure. In that case, an underwriting team may use this information, along with risk scores, to limit their further exposure to insuring properties in that area. This tool provides management with a big-picture view of their risk concentration, which aids them in many decisions, such as the need for additional mitigation steps, reducing their business footprint in a high-risk area or increasing it in a lower-risk area, and identifying the need for changed capital allocations.

Both tools are designed to facilitate data-driven decision-making. As the severity and frequency of natural disasters have increased, and as the use of information technology has expanded, the expansion of these tools is not surprising. More recently, the new artificial intelligence (AI) programs are adding to the development of these two tools.

As a business based on data, statistics, probabilities, etc., the latest tools are impressive in helping us better quantify risk, but, like most statistics, they may have unintended consequences. Sometimes simple solutions lead to additional issues, as many insurance agents in high-risk wildfire areas have learned. Additionally, their use and underlying algorithms are becoming more prone to regulatory oversight. As these tools impact the availability and affordability of residential insurance, this regulatory trend is likely to accelerate.

For areas with high natural catastrophe loss exposures, actuaries have developed more sophisticated "catastrophe modeling" that uses variations of scoring and mapping to identify areas of high risk for their customers. These models are additional tools used by underwriters to measure and price risk in highly exposed areas. Such modeling is usually not transparent to insurance agents and their customers, although recent California and Colorado legislation and regulations are beginning to make them more transparent.

Although it is assumed that these models can correctly predict areas with higher exposure, recent natural disasters suggest that the industry is far from having reliable tools to predict these events. A continuing challenge is to adjust these models to conform to the ever-changing extreme weather conditions to keep them relevant.

While risk scoring and mapping represent significant advances in data-driven underwriting, they face a fundamental paradox: The very catastrophes they aim to predict are becoming less predictable, more severe, and increasingly divorced from historical patterns.

Early Real-World Impacts

Since Colorado has experienced many recent wildfires, it is not surprising that the use of these tools increased as companies sought to understand their wildfire-exposed properties better. While there were some attempts to better manage the wildfire exposure before the 2021 Marshall Fire, this multibillion-dollar loss has changed the whole personal lines insurance landscape in Colorado. Let me share two examples of scoring I witnessed before and after the Marshall Fire.

• The first instance of experiencing risk scores was when I moved from my residential home into a senior cooperative in 2021. When I went to change my policy from an HO-5 to an HO-6, my insurer, which had insured me for 40 years, advised me that it could not issue the new policy because we were in a high-fire zone. I asked my agent about this declination and learned that their new scoring was based on zip codes and recent wildfire activity. This determination surprised me because we knew that the Colorado Springs Fire Department's (CSFD's) mapping of our area did not classify our area as wildfire-prone. We decided to ask our former company contacts to intervene on our behalf, using drone photos of our new residence and other wildfire-specific information. We were successful in our appeal but found the policy could not be issued by the system because there was no way to override the score at that time.
• Fortunately, based on our experience with the Waldo Canyon Fire and Black Forest Fire, we were able to leverage that knowledge to our benefit. On several occasions since then, I have inquired about the current scores at this location and learned that the newer granular scores (specific to our location) continue to rate this location as a medium wildfire risk. We live in a neighborhood on the high prairie, not in the trees nor on the edge of grasslands, and a separate CSFD inspection has confirmed sound mitigation and a lower wildfire risk.

These latter issues continue to evolve among other Colorado agents, especially those near recognized wildland-urban interfaces. All this was happening as insurers realized they needed to adjust their underwriting tactics in response to the negative loss results from the Marshall Fire. To add fuel to the fire (pun intended), large wildfires keep occurring, such as the Lahaina Fire and the Los Angeles Fire, which continue to challenge the insurance industry's ability to predict their frequency and severity.

Risk Scoring and Mapping: Practical Uses

Today's insurers use these tools to aid many of their operations and decisions. Top financial management uses them to accomplish the following.

• Understand the exposures they are willing to write.
• Refine pricing for higher-risk customers.
• Control aggregation of high-risk exposures in their book of business.
• Assist in making reinsurance purchases.
• Allocate capital to minimize financial disruption when catastrophes occur.
• Develop more customized, segmented policies to address high-catastrophe locations.

The typical line departments use these tools during their operations as follows.

• Underwriters use scoring to identify insurable differences between properties down to the actual location or at least the neighborhood level.
• Claims department management can use risk mapping to assist in setting reserves when catastrophes occur and perhaps to guide the allocation of personnel to handle a disaster recovery.
• Marketing and agency departments can use this information to encourage their customers to consider mitigating their disaster exposures or to communicate why there are premium differences between customers located near each other.

Since these tools are widely used, there is little doubt that the insurance industry relies on them across many facets of its business. They are most likely more reliable for standard insurance risk exposures but less helpful for catastrophic disaster decision-making as ever-changing weather science continues to add more variables to their underlying algorithms. The next challenge is introducing greater transparency with their agents and customers, which will raise new questions about the reliability of these scores.

Like all tools, there are pros and cons to their use. As noted earlier, when credit scores were introduced many years ago, they too were met with some resistance. Still, over time, their use was adjusted to make them acceptable to most insurance departments. Since insurance credit scores were like financial credit scores, their use was easier once people understood the relationship between those scores and a person's insurability. One attribute of credit scores is the ability for customers to repair their scores over time.

Whether risk scores will provide customers with the same ability remains a question as they become more transparent to the public. Since risk scores aim to quantify catastrophic risk, the question remains how readily their continued growth will affect the insurance industry, its frontline agents, and American consumers.

The Path Forward

At first glance, risk scores seem like a simple, straightforward way to use data to arrive at a single, easy-to-quantify factor for property risk. Beneath the surface, they face many potential challenges with continued use. Let's look at several larger issues that are likely to evolve with these tools.

From my perspective, one of the larger hurdles is how or if they can adequately assess the impacts of natural catastrophes at a granular level. With each passing year, we learn that the effects of extreme weather and climate change are poorly understood and rapidly changing. Successful modeling will need to adapt to these changes, which may have a limited lifespan until another variable changes. Because the most extreme catastrophes appear to be isolated events, algorithms may lack sufficient data to predict their impact. Consider just these simple examples of the volatility of change.

• Wildfires are supposed to occur in the western United States, but now we are seeing other areas begin to experience these disasters. How do models predict drought and wind, two significant accelerants of wildfire?
• Floods are supposed to occur in historically mapped coastal and fluvial areas, as outlined in the NFIP maps. As we are now continually witnessing, nature does not always follow historical patterns, nor does it produce the same amount of water that initiates these extremely destructive events.
• Tornadoes were long assumed to be an attribute of the "tornado alley" in the United States. Now we see tornadoes occurring in many more areas, and they seem to be shifting eastward and southeastward.
• Convective weather and hail events were historically significant weather events in the central United States and are now moving eastward. As a percentage of total catastrophic losses, these storms are becoming a substantial contributor to the rising total.

The challenge for vendors and actuaries is to continually adapt their models to these rapidly changing exposures and keep them relevant to the insurance mechanism's needs to price its product correctly.

A central purpose of scoring is to quantify risk and allocate anticipated losses to customers most likely to experience them. On the surface, this seems like a worthy objective, but the actual outcomes may differ. For insurance consumers living in designated catastrophe areas, insurance costs may rise precipitously, making adequate coverage either unavailable or unaffordable.

Wildfire-prone states are already facing this problem, which, so far, seems to have no easy solutions. As this occurs, the call for more legislative and regulatory oversight creates an even more challenging environment for insurers. Even though another purpose of scoring is to promote mitigation, the impacts and costs of mitigation will likely pose a significant hurdle for Main Street America. They will require continual education and financial resources to accomplish.

Finally, recent legislative actions in California and Colorado aim to make scoring, mapping, and modeling more transparent, which will likely lead to considerable delays in implementing some rate changes, particularly those affecting people in highly exposed catastrophe areas and economically challenged areas. Although the impacts of these regulatory changes are unknown due to their recent enactment, there is little doubt that there will be greater friction in marketing, pricing, and risk management for insurers and their agents alike.

Colorado Legislative and Regulatory Response

Recent legislative action in Colorado, HB25-1182 Risk Model Use in Property Insurance Policies, aims to make scoring, mapping, and modeling more transparent. This legislation requires a variety of transparent communications regarding risk modeling and scoring to the insurance commissioner and the customers. Here are four key requirements for insurance companies to comply with.

• Submit available data concerning the models and scoring methods to the commissioner as part of the insurer's rate filings.
• Verify that certain risk factors are included in their wildfire risk model, catastrophe model, or combination of models.
• Notify the insurance customers within 15 days of their initial application for insurance, and upon all renewals, their risk scores and what mitigation discounts are available to them on standardized forms.
• Make available on the insurer websites a clear description of risk scoring and models, and share how these scores may be improved with either property-specific or community mitigation.

Although the final regulation has not yet been finalized, it is evident that this legislative initiative will change the way insurers and their agents communicate information about these items to both regulators and customers. The legislation even requires contesting the scores. The final impact of this regulation is unknown at the time of this article, but the way property insurance is written and priced will change.

Another key intent of this legislation is to encourage greater community and insurance consumer mitigation efforts to control the long-term costs of property insurance for Colorado residents. Indeed, this is a good public policy goal. However, it is not yet clear what the cost-benefit of these risk-reduction activities is for offsetting the ever-increasing costs of wildfires and other natural disasters.

Like any significant change, this regulation will likely encounter some hurdles as it is implemented. As the insurance industry and public policy experiment with methods to control the rising insurance costs from natural catastrophes, insurance professionals need to continually seek ways to better quantify and mitigate the costs associated with these growing climate-related costs.

Takeaways

Indeed, risk scoring seems an elegant way to quantify risk and price insurance products. However, it is unlikely to be an easy fix to a very complex problem. The biggest hurdle is recognizing the problem and starting to work collaboratively to find solutions. Old ways are unlikely to address this evolving worldwide issue, so all insurance professionals and consumers will need to work to find incremental solutions until either one solution or a combination of many solutions achieves better approaches to resolving these issues.

Because we are dealing primarily with "natural perils," it is expected that this road will be long and bumpy, especially for those insurance professionals working in present-day high-catastrophe areas. Those living outside these areas need to stay vigilant because history and science suggest they will be affected at some point in the future.

As we know, challenges bring opportunities. As a retired independent insurance agent, I believe the relevance of agents will only increase in the near and long term. While some fear being replaced by AI or by Internet insurance salespeople, the importance of personal guidance in finding solutions and helping individual customers navigate this new, evolving insurance reality is greater than you could ever imagine. To facilitate survival, today's insurance agents need to retool their education to understand the changing dynamics of natural catastrophes and to collaborate with their community's first responders to have open conversations and share ideas for handling these increasingly frequent and severe insurance events. It is time to stop denying the growing problem and become part of its solution.